https://esa.github.io/pagmo_plugins_nonfree/cpp_snopt7.html
This repository is a tutorial on how to install and run SNOPT for optimization using pagmo (https://github.com/esa/pagmo2). A conda environment is used to install both pagmo and pagmo_plugins_nonfree (https://github.com/esa/pagmo_plugins_nonfree).
Works with Ubuntu 20.04 and miniconda for Python 3.9 but is yet to be tested on a fresh installation of the operating system.
(Very much inspired by https://github.com/tudat-team/tudat-bundle)
1. Request a "Non-U.S.-based" 3-month trial license for SNOPT7 from https://ccom.ucsd.edu/~optimizers/downloads/.
The license should be requested for Linux and should have the Fortran libraries.
git clone http://github.com/castanhas98/install-snopt-tutorial
cd install-snopt-tutorial
git submodule update --init --recursive
According to esa/pagmo_plugins_nonfree#2, a very specific commit of the snopt-interface (https://github.com/snopt/snopt-interface/commit/76b166ecdf5c55a3289ce0f849d8d3d101954a22). However, this is handled but the command above, so there is nothing else to be done in that regard.
sudo apt-get update
sudo apt-get upgrade
sudo apt-get install autoconf build-essential libtool gfortran
5. Configure the snopt-interface (taken from esa/pagmo_plugins_nonfree#2 (comment)).
cd snopt-interface
./autogen.sh
./configure
make snopt_c
After the last command, an error similar to the following will be displayed:
/usr/bin/ld: cannot find -lsnopt7
collect2: error: ld returned 1 exit status
make: *** [Makefile:137: lib/libsnopt7_c.la] Error 1
This is normal and will be solved in the coming steps. The command needs to be ran anyways to ensure the correct creation of the /install-snopt-interface/snopt-interface/lib directory and of some of the contents in the /install-snopt-interface/snopt-interface/src directory.
Some amount of time after requesting the license, an email will be sent to the registered email address. That email will contain both the license (snopt7.lic) and a link to the downloads, which looks like http://ccom.ucsd.edu/~optimizers/downloads/software/academic?id=XXXXXXXXXXXX. The "Fortran (only) Libraries" are the ones that should be downloaded.
The libsnopt7.zip file should be extracted into /install-snopt-tutorial/snopt-interface/lib. Note that this directory is only generated after Step 5 is carried out.
Similarly to Step 5, the commands below are ran insed the /install-snopt-tutorial/snopt-interface directory.
make install snopt_c
For a temporary addition, that exists while the current terminal exists and that needs to be ran every time a new terminal is opened:
export LD_LIBRARY_PATH=`realpath ./lib`
or
export LD_LIBRARY_PATH=/absolute/path/to/snopt-interface/lib`
In order to avoid having to do this every time a new terminal window is opened, one can create a new .conf file under /etc/ld.so.conf.d:
sudo nano /etc/ld.so.conf.d/snopt.conf
And writing /absolute/path/to/snopt-interface/lib inside, which can be obtained through the realpath command shown above. Ctrl + X, Y, Enter to save and exit. Afterwards, run:
sudo ldconfig
to update the system with the new libraries.
The snopt7.lic file that is sent in the email must be saved somewhere in the computer. Assuming that it is saved under /install-snopt-tutorial:
cd .. # to get back to the /install-snopt-tutorial directory
export SNOPT_LICENSE=`realpath ./snopt7.lic`
Again, a more permanent solution by adding a line to the /etc/environment file. Open it by doing:
sudo nano /etc/environment
and add a new line to the file with the following:
SNOPT_LICENSE="/absolute/path/to/snopt7.lic"
where /absolute/path/to/snopt7.lic is the absolute path to the snopt7.lic file. Again, Ctrl + X, Y, Enter to save and exit. The computer should be restarted after this step.
sudo apt-get install libgfortran4
In case libgfortran4 is not found, run the following commands:
sudo add-apt-repository universe
sudo apt-get update
sudo apt-get install libgfortran4
After the steps above are taken, one can create the conda environment in which pagmo and tudat will be installed.
Assuming the current directory is /install-snopt-tutorial:
conda env create -f environment.yaml
conda activate snopt-pagmo-env
When using snopt7 as a pagmo::algorithm, the path (absolute or relative) to the C library (the libsnopt7_c.so file under /install-snopt-tutorial-snopt-interface/lib) that was built in Step 7 must be present when the pagmo::algorithm is declared.
Even though the examples work as they are, it is advised to change the lines of the .cpp files in the repository that look like the one below
algorithm algo(ppnf::snopt7(false, "./snopt-interface/lib/libsnopt7_c.so", 6u));to
algorithm algo(ppnf::snopt7(false, "/absolute/path/to/install-snopt-tutorial/snopt-interface/lib/libsnopt7_c.so", 6u));where /absolute/path/to/install-snopt-tutorial/snopt-interface/lib/libsnopt7_c.so is the absolute path to the libsnopt7_c.so file.
This step requires CMake to be installed. It can be downloaded from https://cmake.org/download/ and installed by following the instructions under "Latest Release" in the website.
Inside /install-snopt-tutorial, run:
bash build.sh
Example 1:
./build/example_1
Which outputs:
(snopt-pagmo-env) $ ./build/example_1
Initial Population:
Problem name: Multidimensional Rosenbrock Function
C++ class name: pagmo::rosenbrock
Global dimension: 30
Integer dimension: 0
Fitness dimension: 1
Number of objectives: 1
Equality constraints dimension: 0start
Upper bounds: [10, 10, 10, 10, 10, ... ]
Has batch fitness evaluation: false
Has gradient: true
User implemented gradient sparsity: false
Expected gradients: 30
Has hessians: false
User implemented hessians sparsity: false
Fitness evaluations: 1
Gradient evaluations: 0
Thread safety: constant
Population size: 1
List of individuals:
#0:
ID: 4880189657833503230
Decision vector: [3.89267, 7.66399, 7.86918, 7.70878, 4.35346, ... ]
Fitness vector: [3.34052e+06]
Champion decision vector: [3.89267, 7.66399, 7.86918, 7.70878, 4.35346, ... ]
Champion fitness: [3.34052e+06]
Final Population:
Problem name: Multidimensional Rosenbrock Function
C++ class name: pagmo::rosenbrock
Global dimension: 30
Integer dimension: 0
Fitness dimension: 1
Number of objectives: 1
Equality constraints dimension: 0
Inequality constraints dimension: 0
Lower bounds: [-5, -5, -5, -5, -5, ... ]
Upper bounds: [10, 10, 10, 10, 10, ... ]
Has batch fitness evaluation: false
Has gradient: true
User implemented gradient sparsity: false
Expected gradients: 30
Has hessians: false
User implemented hessians sparsity: false
Fitness evaluations: 1
Gradient evaluations: 0
Thread safety: constant
Population size: 1
List of individuals:
#0:
ID: 4880189657833503230
Decision vector: [1, 1, 1, 1, 1, ... ]
Fitness vector: [1.08934e-16]
Champion decision vector: [1, 1, 1, 1, 1, ... ]
Champion fitness: [1.08934e-16]
Example 2:
./build/example_2
Which outputs:
It probably will not give exactly the same output since the seed used to generate the population of each island is not set. It is important, however, that the final archipelago displays fitness function values much lower than the original ones.
(snopt-pagmo-env) $ ./build/example_2
[4.18693e+06]
[5.87389e+06]
[4.50279e+06]
[4.19484e+06]
[2.43854e+06]
[4.16832e+06]
[2.76975e+06]
[3.65566e+06]
[4.08998e+06]
[2.6937e+06]
[2.23983e+06]
[3.26988e+06]
[5.35522e+06]
[4.2625e+06]
[1.37748e+06]
[4.33746e+06]
Fitness function values after the optimization:
[4.18693e+06]
[3.98662]
[5.61898e-15]
[4.2837e-15]
[3.98662]
[3.12364e-15]
[2.84757e-15]
[4.30476e-17]
[8.93509e-16]
[1.97794e-15]
[1.65367e-14]
[1.85848e-16]
[3.98662]
[8.02854e-16]
[1.95138e-16]
[1.88996e-15]
Sometimes, the following error appears:
At line 71 of file src/sn01tri.f (unit = 10, file = 'fort.10')
Fortran runtime error: End of file
Segmentation fault (core dumped)
Which is believed to be from pagmo's or SNOPT's side, but further troubleshooting needs to be done. However, running the executable again until it works seems to be a decent impromptu workaround.
Example 2 is very similar to the code provided in https://esa.github.io/pagmo_plugins_nonfree/quickstart.html. It must be said that that example performs the evolution in a pagmo::archipelago instead of in a pagmo::population as it is done in examples 1 and 3.
Example 3 (using Tudat):
./build/example_3
Which outputs:
(snopt-pagmo-env) $ ./build/example_3
Some requirements:
- A trial version license from https://ccom.ucsd.edu/~optimizers/downloads/
anacondaorminicondagitautoconf,build-essentialandlibtoolpackages- g++ (pre-installed with Ubuntu 20.04)
- A Fortran compiler, we used gfortran
cmake
Some thoughts:
- Add the path with libsnopt7.so to LD_LIBRARY_PATH or to /etc/ld.so.conf.d/ create a file like snopt.conf and write the directory there. Path to directory with file, not path to file. Absolute paths from root /, not from ~
- Add SNOPT_LICENSE to /etc/environment